Methodology

The initial matching pass uses an 8-signal scoring algorithm to identify the most likely OpenAlex author profile for each member.

Signals Used:

1. Name match: Exact, fuzzy, and initial-based matching (Levenshtein distance, Jaro-Winkler)
2. Institution affiliation: Australian/NZ institution presence in OpenAlex author record
3. Country filter: Author must have at least one AU/NZ affiliated work
4. Topic relevance: Rehabilitation medicine keywords in publication topics
5. ORCID cross-reference: If member has known ORCID, verify it matches
6. Publication count plausibility: Clinical researchers typically have 1-200 papers
7. Career timeline: Publication years should be plausible given career stage
8. Disambiguation signals: For common names, require multiple corroborating signals

Confidence Tiers:

• HIGH: 3+ corroborating signals, no conflicting evidence
• REVIEW: Partial match, some ambiguity, or common name
• NOT_FOUND: No matching author profile identified

Cross-validates Round 1 matches against ORCID and Crossref APIs. Identifies and removes incorrect matches.

Verification Steps:

1. Query OpenAlex author record: verify name, institution, works count
2. Query ORCID API: check if ORCID record exists and matches name/affiliation
3. Query Crossref: search for publications by name, verify DOI overlap
4. Geographic validation: confirm at least one AU/NZ affiliation in works
5. Topic validation: check if publication topics align with rehabilitation medicine
6. Flag mismatches: if record shows different country or unrelated field, mark as REMOVED

Addresses the "author fragmentation" problem where OpenAlex creates multiple separate profiles for the same person.

The Problem:

OpenAlex often splits a single researcher into multiple author entities when they publish infrequently, change institutions, lack ORCID linkage, or have no coauthor overlap between papers.

The Solution: Works-Level Search

1. Search OpenAlex works endpoint by raw_author_name
2. For each work found, extract the associated author_id
3. Apply geographic filter: only retain works with AU/NZ institutional affiliations
4. Apply name matching: verify author name on the work matches the member name
5. Cluster all discovered author IDs into a unified profile
6. Merge publication lists and recompute metrics (H-index, citations, FWCI)

Addresses a critical bug in Round 3 where name-matched profiles were incorrectly merged without disambiguation, inflating publication counts for members with common names.

Root Cause:

Round 3's works-level search found ALL OpenAlex author profiles matching a member's name and merged their publications without verifying they were the same person. For common names (e.g., "Chris Chan" matched 64 profiles, "Michael Smith" matched 29), this produced impossibly high publication counts (5,000+).

Fix Applied:

1. Fetched authoritative metrics directly from each member's verified primary OpenAlex ID
2. Replaced inflated h-index, citation count, and publication count with primary-ID-only values
3. Filtered publication lists to cap at the correct count from the primary ID
4. Added "Suspicious Merge" flag for non-HIGH-confidence members with >5 merged IDs
5. HIGH-confidence members (verified in R1/R2) are exempt from suspicious flags — their primary ID was already validated